Clock syncrhonisation limits in a cluster?

qcope

We have a cluster of eight nodes, where clock synchronisation through ntpd has not been working well and clocks have been getting out of sync. I know the fix is to fix the clocks, we are on it.... but does anyone know what the absolute limits are for discrepancies in the system clocks, of nodes in a cluster?

Jim_Knicely

According to the vnetperf optimization tool, it is recommended to keep the clock skew < 1 second.

Example:

[dbadmin@SE-Sandbox-43-node1 ~]$ /opt/vertica/bin/vnetperf --hosts 192.168.61.227,192.168.61.228,192.168.61.229 latency

The maximum recommended rtt latency is 2 milliseconds. The ideal rtt latency is 200 microseconds or less. It is recommended that clock skew be kept to under 1 second.
test              | date                    | node             | index | rtt latency (us)  | clock skew (us)
-------------------------------------------------------------------------------------------------------------------------
latency           | 2019-03-07_11:31:15,76  | 192.168.61.227   | 0     | 50                | 1
latency           | 2019-03-07_11:31:15,76  | 192.168.61.228   | 1     | 91                | -2179
latency           | 2019-03-07_11:31:15,76  | 192.168.61.229   | 2     | 104               | 1148

JSON results available at: ./results.2019-03-07_11:31:13,608.json

See:
https://www.vertica.com/docs/9.2.x/HTML/Content/Authoring/InstallationGuide/scripts/vnetperf.htm

qcope

Thanks for this! Sorry, read it at the time and it was useful and then got caught up in other problems.